Optimising Shopify for AI agents: what actually helps

When a shopper asks an AI assistant to “find a fragrance-free cream for sensitive skin under €20”, the agent doesn’t browse stores like a person. It reads structured data: category, properties, price, availability and store policies. On Shopify, the main source of that data is Shopify Catalog.

The good news is that most of the work happens in the admin and in product data, not in code. The bad news is that none of it guarantees visibility. Shopify says so plainly: being in Catalog doesn’t mean a product will appear in a specific AI answer, because each channel controls its own ranking and presentation.

1. Start with what Shopify Catalog sees

In the admin, open Sales channels > Agentic, choose Shopify Catalog under Sources and click Review product data mapping. Each field shows the share of products with a value, and “Fix … products” takes you straight to the gaps (mapping documentation). It’s the fastest AI-channel audit you have, and it’s free.

2. Categories and properties as data, not sentences

Give every product the most specific category in the Shopify taxonomy. Shopify’s guide to agent-ready product data uses “men’s insulated winter boots” rather than “footwear” as the example. The product category can’t be remapped in Catalog mapping, so it has to be right on the product itself.

Properties shoppers decide on belong in fields and metafields: material, size, colour, ingredients, suitability for sensitive skin. An agent looking for “fragrance-free” needs that as a value, not a word buried in a paragraph.

3. Write titles and descriptions literally

The title should say what the product is. The description should explain in natural language who it’s for, what it’s made of and how it differs. Shopify’s article on agentic shopping recommends factual copy without vague marketing phrases or keyword lists. Don’t bulk-fill empty fields with generated claims you haven’t verified: the agent will repeat them to the shopper as fact.

4. Variants belong to one product

If the same item comes in several versions, group them under one product. Otherwise agents may treat them as different things and answer wrongly, for example about available colours. If you keep variants as separate products for good reasons, such as their own URLs, use custom variant grouping in Catalog mapping. Two variants with the same option combination can’t be grouped, and custom grouping overrides the Combined Listings app.

A real example: in the Bohemian Coffee House store, the light and dark roast of the same coffee are deliberately two products with a shared switch. That’s right for the website; for the catalog it’s worth grouping them.

5. Don’t rely on the theme

Information that exists only in a Liquid template, in JavaScript or in section text isn’t visible to agents through the catalog. Shopify Enterprise points this out explicitly. Move it into product fields or metafields and let the mapping build the title or description from them. There’s a bonus for the site: the same data can then drive filters, comparisons and the theme.

6. Price, stock and policies must match

Agents work with live price and availability, and a mismatch between stock and the site ends in a lost purchase. Since 9 Oct 2026 Shopify shares compare-at prices as list price for most stores (changelog). In the EU, an announced discount must be based on the lowest price of the previous 30 days, so a struck-through price has to reflect reality. Discounts applied only in the cart don’t show in the catalog price.

Terms of service, privacy and return policies are required for AI channels. Describe shipping, delivery times and exchanges concretely, and review the facts in the Knowledge Base (Sales channels > Agentic > Sources). Publish only what any shopper may see.

7. Test and measure

In Sales channels > Agentic, try Search preview with queries your customers actually use: product type, use, property, size. It’s Catalog’s raw output, not the ranking in ChatGPT, but it shows missing data quickly. Track AI channel performance in the Agentic reports. With Google direct checkout, Google Analytics and custom pixels don’t fire (Google channel requirements), so compare orders in the admin, not just GA4.

What merchants and agencies recommend

Beyond the official documentation, these points keep coming up in merchant discussions and Shopify podcasts over the last weeks. They aren’t Shopify requirements, but sensible practice:

• Identifiers: GTIN or EAN and brand for products that have them. For the Google channel, which draws on Merchant Center, they’re a common requirement.
• Images: several product photos from different angles, with descriptive alt text.
• Reviews: real ratings and review text help in comparisons. Never invent them.
• Categories and attributes first: they’re the most structured part of what Shopify sends to agents.
• Crawler access: if you’ve edited robots.txt, check you aren’t blocking AI search crawlers.

What to avoid

• Apps promising “first place in ChatGPT”. The channel decides the ranking, not an app in your store.
• Bulk-generated descriptions with claims about ingredients, certification or effects you can’t back up.
• Key details only in images, JavaScript-loaded tabs or PDFs.
• Different prices on the site, in the catalog and in Merchant Center.

Checklist

• Field coverage in Catalog mapping is close to 100%
• Every product has the most specific category available
• Deciding properties live in fields or metafields, not only in text
• Variants are grouped under one product or via custom variant grouping
• Prices, compare-at prices and stock match reality and the law
• Store policies and Knowledge Base facts are complete and public
• Search preview returns the right products for real queries